Liver Segmentation in CT Data:
A Segmentation Refinement Approach*

Reinhard Beichel1, Christian Bauer2, Alexander Bornik2,
Erich Sorantin3, and Horst Bischof2

1 Dept. of ECE & Internal Medicine, The University of Iowa, USA
2 Inst. for Computer Graphics and Vision, Graz University of Technology, Austria
3 Department of Radiology, Medical University Graz, Austria

* In Proceedings of "3D Segmentation in The Clinic: A Grand Challenge",
T. Heimann, M. Styner, and B. van Ginneken, Eds., 2007, pages 235-245, ISBN 978 0 643 09523 6.

Abstract

Liver segmentation is an important prerequisite for planning of surgical interventions like liver tumor resections. For clinical applicability, the segmentation approach must be able to cope with the high variation in shape and gray-value appearance of the liver. In this work we present a novel segmentation scheme based on a true 3D segmentation refinement concept utilizing a hybrid desktop/virtual reality (VR) user interface. The method consists of two main stages. First, an initial segmentation is generated using graph cuts. Second, an interactive segmentation refinement step allows a user to fix arbitrary segmentation errors. We demonstrate the robustness of our method on ten contrast enhanced liver CT scans. Our segmentation approach copes successfully with the high variation found in patient data sets and allows to produce accurate segmentations in a time-efficient manner.

Motivation

Automated liver segmentation methods frequently fail to deliver good segmentation results, especially in case of diseased livers. For clinical application, liver segmentation must be capable of handling all possible cases in a time-efficient manner. To tackle this problem, we propose a novel refinement approach to 3D liver segmentation.

Method

The proposed approach to liver segmentation consists of two main stages: initial segmentation and interactive segmentation refinement. As input for the first stage, a CT volume and one or more start regions, marking liver tissue, are used. The segmentation is then generated using a graph cut (GC) approach. In addition, a partitioning of the segmentation and the background into volume chunks is derived from edge/surface features calculated from the CT volume. These two types of output are passed on to the second stage which allows for the correction/ refinement of segmentation errors remaining after the first stage. Refinement takes place in two stages. First, volume chunks can be added or removed during the chunk-based refinement (CBR) stage. This step is usually very fast, and the majority of segmentation errors occurring in practice can be fixed or at least significantly reduced. Second, a mesh-based refinement (MBR) allows to correct arbitrary errors. For this purpose, the binary segmentation is converted to a simplex mesh which is deformed during MBR by utilizing various tools. Each of the refinement steps is facilitated using interactive VR-enabled tools for true 3D segmentation inspection and refinement, allowing for stereoscopic viewing and true 3D interaction. Since the last stage of the refinement procedure is mesh-based, a voxelization method is used to generate a labeled volume

Fig. 1: System overview.




Fig. 2: Illustration of the system components and the refinement stages. Both, chunk-based (Fig. 2.B) and mesh-based refinement (Fig. 2.C) utilize a hybrid desktop/VR user interface shown in Fig. 2.A.


Results

Our approach to liver segmentation was evaluated ten test CT data sets (see [1,2] for details). The performance measures and scores clearly show the effectiveness of the segmentation refinement concept: metrics and scores improve with each refinement stage, while the required user interaction is low compared to a manual segmentation (~60 to 70 minutes). Using the segmentation refinement approach, a high segmentation quality (mean average distance of less than 1 mm) can be achieved. In addition, the interaction time needed for refinement is approximately 6.5 minutes on average. Thus, the presented refinement concept is well suited for clinical application in the context of liver resection planning.





Fig. 3: Summary of results for all three processing stages. See [1] for details.


References:

[1] R. Beichel, C. Bauer, A. Bornik, E. Sorantin, and H. Bischof. "Liver segmentation in CT data: A segmentation refinement approach". In Proceedings of 3D Segmentation in The Clinic: A Grand Challenge, T. Heimann, M. Styner, and B. van Ginneken, Eds., 2007, pages 235-245, ISBN 978-0-643-09523-6.

[2] T. Heimann, M. Styner, and B. van Ginneken, "3D segmentation in the clinic: A grand challenge", in 3D Segmentation in The Clinic: A Grand Challenge, T. Heimann, M. Styner, and B. van Ginneken, Eds., 2007, ISBN 978-0-643-09523-6.


Contact: reinhard-beichel(at)uiowa(dot)edu